XYZZY
Self-hosted multiplayer AI workspace for technical decision-making with tamper-evident audit trails and governed agent orchestration. Built in Python with persistent event sourcing and real-time WebSocket collaboration.
🔗 Visit XYZZY
Description
Picture a team of AI assistants working together on a complex decision — like reviewing a contract or designing a system — and being able to see, later, exactly which piece of evidence led to which conclusion, with nothing quietly lost or altered along the way. XYZZY is built for that: a shared workspace where multiple AI agents collaborate under human oversight, and every step leaves a tamper-evident trail you can audit.
XYZZY is a self-hosted, open-source (Apache 2.0) multiplayer AI workspace written in Python with a FastAPI backend, SQLite by default (optional Redis for multi-process scaling), and real-time WebSocket collaboration. It orchestrates multiple AI agents with human-in-the-loop approval gates, persists a hash-chained event-sourced audit trail, versions artifacts, and links decisions to the underlying evidence through a structured ontology. It supports local models via Ollama or LM Studio as well as the OpenAI API, and ships with 954 passing tests.
💬 Our review
The short version: XYZZY targets a specific, underserved need — multi-agent AI collaboration that produces an audit trail regulators or auditors can actually trust, not just a chat log.
Compared to AutoGen, LangGraph, or CrewAI — which focus on getting agents to coordinate — XYZZY's differentiator is the compliance layer: immutable, hash-chained event sourcing and an evidence ontology that ties every output back to its source, which matters a lot more in regulated or research environments than in a typical startup's internal tooling. The tradeoff is that this governance layer adds real complexity: you're running a Python/FastAPI service, thinking about SQLite-vs-Redis scaling, and learning the evidence-ontology concepts, none of which are needed if you just want agents to get a task done fast. For compliance-heavy teams that need to prove how an AI-assisted decision was reached, XYZZY is worth the setup cost; for quick prototyping or a small team without audit requirements, a lighter framework like CrewAI will get you moving faster.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Gratuit, licence Apache 2.0, auto-hébergé (SQLite par défaut, Redis optionnel pour la montée en charge).
Pros
Pistes d'audit immuables idéales pour la conformité et les exigences réglementaires
Vraie coordination multi-agents avec portes d'approbation humaine
Entièrement auto-hébergeable, aucune dépendance externe requise
Riche ontologie de preuves reliant décisions et résultats
Cons
Backend Python/FastAPI qui peut demander une charge DevOps au déploiement
Courbe d'apprentissage sur les concepts d'ontologie de preuves et de gouvernance
Écosystème plus restreint que les frameworks LLM grand public
SQLite par défaut, à surveiller pour les grandes équipes